Disambiguation of semantic types in complex noun phrases for extracting candidate terms

摘要

Mapping concepts from medical resources to structured medical documents is a prerequisite for many automatic document processing tasks. These resources are characterised by an abundance of material to represent any given concept. Moreover, the resources may include ambiguous terms in unstructured form that lead to distorted results in automating biomedical text mining. This paper is an exploratory study on disambiguation of semantic types for extracting a structured taxonomy from unstructured reports. Specifically, the terms that will be disambiguated are terms that have more than one semantic type in the Unified Medical Language System (UMLS) Metathesaurus. We suggest a word sense disambiguation algorithm that utilises the UMLS is-a hierarchy, augmented with a higher level representing semantic groups, as a knowledge base. The purpose is to explore all possible commonalities to classify simple or composed candidate terms with the Nearest Common Kinship (NCK). Experiments with the training corpora provide encouraging results.